Use Python on Machine Learning

There are a number of factors that make you happy! The World Happiness Report numerates the happiness index by country with a number of factors considered. We utilized Machine Learning techniques to conduct further analysis to explore the level of impacts. After cleaning and training our dataset, we ran a multi-regression analysis and created a happiness formula to better explain the correlations among the happiness factors and happiness score. We analyzed/identified that seven of these factors have a more significant impact on happiness score. We discovered that freedom and life expectancy were the factors that had the biggest impact, and generosity has a surprisingly smaller effect on happiness score. The correlations among those factors are all positive as well, which explains the existence of ripple effects These possible gaps among countries on the happiness index can be expected to vary depending on the inputs/factors involved.

In [1]:
# Update sklearn to prevent version mismatches
  # !conda install scikit-learn
  # !conda update scikit-learn
  # !conda install joblib 
  # !conda update joblib
In [2]:
%matplotlib inline
  import pandas as pd
  import numpy as np
  import matplotlib.pyplot as plt

Read the CSV and Perform Basic Data Cleaning

In [3]:
# Drop the null columns where all values are null
  df = pd.read_csv("whd-2015-19.csv")
  df = df.dropna(axis='columns', how='all')
In [4]:
new_df_w_continent=df.drop(['country','happiness_rank','year'],axis=1)
  new_df_w_continent = new_df_w_continent.apply(lambda x: x.fillna(0),axis=0)
In [5]:
new_df=df.drop(['country','happiness_rank','continent','year'],axis=1)
  new_df = new_df.apply(lambda x: x.fillna(0),axis=0)
  new_df.head()
Out[5]:
happiness_score gdp_per_capita family life_expectancy freedom generosity government_corr social_support
0 7.587 1.39651 1.34951 0.94143 0.66557 0.29678 0.41978 0.0
1 7.561 1.30232 1.40223 0.94784 0.62877 0.43630 0.14145 0.0
2 7.527 1.32548 1.36058 0.87464 0.64938 0.34139 0.48357 0.0
3 7.522 1.45900 1.33095 0.88521 0.66973 0.34699 0.36503 0.0
4 7.427 1.32629 1.32261 0.90563 0.63297 0.45811 0.32957 0.0
In [6]:
new_df.describe()
Out[6]:
happiness_score gdp_per_capita family life_expectancy freedom generosity government_corr social_support
count 782.000000 782.000000 782.000000 782.000000 782.000000 782.000000 782.000000 782.000000
mean 5.379018 0.916047 0.595221 0.612416 0.411091 0.218576 0.125275 0.483171
std 1.127456 0.407340 0.544504 0.248309 0.152880 0.122321 0.105844 0.622930
min 2.693000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000
25% 4.509750 0.606500 0.000000 0.440183 0.309768 0.130000 0.054000 0.000000
50% 5.322000 0.982205 0.698400 0.647310 0.431000 0.201982 0.090905 0.000000
75% 6.189500 1.236187 1.082465 0.808000 0.531000 0.278832 0.155861 1.174000
max 7.769000 2.096000 1.610574 1.141000 0.724000 0.838075 0.551910 1.644000

Create a Train Test Split

In [7]:
X = new_df.drop("happiness_score", axis=1)
  y = new_df.happiness_score
  print(X.shape, y.shape)
  X
  
(782, 7) (782,)
Out[7]:
gdp_per_capita family life_expectancy freedom generosity government_corr social_support
0 1.39651 1.34951 0.94143 0.66557 0.29678 0.41978 0.000
1 1.30232 1.40223 0.94784 0.62877 0.43630 0.14145 0.000
2 1.32548 1.36058 0.87464 0.64938 0.34139 0.48357 0.000
3 1.45900 1.33095 0.88521 0.66973 0.34699 0.36503 0.000
4 1.32629 1.32261 0.90563 0.63297 0.45811 0.32957 0.000
... ... ... ... ... ... ... ...
777 0.35900 0.00000 0.61400 0.55500 0.21700 0.41100 0.711
778 0.47600 0.00000 0.49900 0.41700 0.27600 0.14700 0.885
779 0.35000 0.00000 0.36100 0.00000 0.15800 0.02500 0.517
780 0.02600 0.00000 0.10500 0.22500 0.23500 0.03500 0.000
781 0.30600 0.00000 0.29500 0.01000 0.20200 0.09100 0.575

782 rows × 7 columns

In [8]:
y
Out[8]:
0      7.587
  1      7.561
  2      7.527
  3      7.522
  4      7.427
         ...  
  777    3.334
  778    3.231
  779    3.203
  780    3.083
  781    2.853
  Name: happiness_score, Length: 782, dtype: float64

Train test_split to create training and testing data

In [9]:
from sklearn.model_selection import train_test_split
  X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=42)
In [10]:
#creating the model using LinearRegression
  from sklearn.linear_model import LinearRegression
  model_n = LinearRegression()
In [11]:
# Fit the model to the training data and calculate the scores for the training and testing data 
  model_n.fit(X_train, y_train)
  training_score = model_n.score(X_train, y_train)
  testing_score = model_n.score(X_test, y_test)
  print(f"Training Score: {training_score}")
  print(f"Testing Score: {testing_score}")
  
Training Score: 0.7804958836281554
  Testing Score: 0.7255338949205835
In [12]:
print("Intercept", model_n.intercept_)
  
Intercept 2.1709104274842916
In [13]:
print("Coefficients", model_n.coef_)
  
Coefficients [0.95038985 0.83456657 1.15166485 1.59090688 0.21018595 1.03319443
   0.65508653]
In [14]:
coef = zip(X.columns, model_n.coef_)
  coef_df = pd.DataFrame(list(zip(X.columns, model_n.coef_)), columns=["Features","Coefficients"])
  coef_df
Out[14]:
Features Coefficients
0 gdp_per_capita 0.950390
1 family 0.834567
2 life_expectancy 1.151665
3 freedom 1.590907
4 generosity 0.210186
5 government_corr 1.033194
6 social_support 0.655087

Regression Equation

Happiness Score= 2.1709104274842916+ 0.950390gdp_per_capita +0.655087social_support +1.151665life_expectancy + 1.590907freedom +0.210186generosity + 1.033194government_corr +0.834567*family

In [15]:
# Plot the Residuals for the Training and Testing data
  plt.scatter(model_n.predict(X_train), model_n.predict(X_train) - y_train, c="blue", label="Training Data")
  plt.scatter(model_n.predict(X_test), model_n.predict(X_test) - y_test, c="orange", label="Testing Data")
  plt.legend()
  plt.hlines(y=0, xmin=y.min(), xmax=y.max())
  plt.title("Residual Plot")
Out[15]:
Text(0.5, 1.0, 'Residual Plot')
In [16]:
import seaborn as sns
In [17]:
#number of variables for heatmap
  k = 7
  corrmat = new_df.corr()
  cols = corrmat.nlargest(k, "happiness_score")["happiness_score"].index
  cm = np.corrcoef(new_df[cols].values.T)
  f, ax = plt.subplots(figsize=(20, 12))
  sns.set(font_scale=1.4)
  hm = sns.heatmap(cm, cbar=True, annot=True, square=True, fmt='.2f', annot_kws={'size': 12},
                   yticklabels=cols.values, xticklabels=cols.values,vmax=1, vmin=0, cmap='YlGnBu')
  hm.set_ylim([7,0])
  g0 = plt.show()
In [18]:
f.savefig('sns_heatmap.png')
In [19]:
# with regression
  g1 = sns.pairplot(new_df, kind="reg")
In [20]:
g1.savefig('sns_pairplot1.png')
In [21]:
 # without regression
  g2 = sns.pairplot(new_df, kind="scatter")
  # plt.show()
In [22]:
g2.savefig('sns_pairplot2.png')
In [23]:
g3 = sns.pairplot(new_df, diag_kind="kde")
In [24]:
g3.savefig('sns_pairplot3.png')
In [25]:
g4 = sns.pairplot(new_df, diag_kind="kde", markers="+",
                   plot_kws=dict(s=50, edgecolor="b", linewidth=1),
                   diag_kws=dict(shade=True))
In [26]:
g4.savefig('sns_pairplot4.png')
In [27]:
g5 = sns.PairGrid(new_df)
  g5.map_upper(plt.scatter)
  g5.map_lower(sns.kdeplot)
  g5.map_diag(sns.kdeplot, lw=3, legend=False);
In [28]:
g5.savefig('sns_pairplot5.png')
In [29]:
g6 = sns.pairplot(new_df_w_continent, hue="continent", palette="Set2", diag_kind="kde", height=2.5)
In [30]:
g6.savefig('sns_pairplot6.png')
In [31]:
g7 = sns.PairGrid(new_df_w_continent, hue="continent", palette="Set2", height=2.5)
  g7.map_upper(plt.scatter)
  g7.map_lower(sns.kdeplot)
  g7.map_diag(sns.kdeplot, lw=3, legend=False);
  g7 = g7.add_legend()
In [32]:
g7.savefig('sns_pairplot7.png')